Analyzing and detecting social spammers with robust features
Bibliographic record
Abstract
The rapid growth of online social networks has attracted an increasing number of social spammers.Spammers gain profits by posting various content such as rumors and malwares.These behaviors greatly compromise social networks' privacy and security, and endanger the whole network community.In the last few years, researchers have proposed a number of spam detection strategies.However, spammers become harder to be detected as they constantly evolve to evade detection by emulating legitimate users and hiding spam patterns.Many detection methods become ineffective.In this thesis, we aim to design spam detection methods using features that are resilient to evolving spammers.To achieve this goal, we first conduct an in-depth analysis on different properties of user accounts.We study the Twitter accounts and extract four kinds of features: profile-based, content-based, community-based and time-based features.By analyzing evasion techniques used by current spammers, we find that the commonly-used profile and content-based features are not effective enough to uncover cunning spammers.This is because these features can be easily emulated by spammers.To tackle this issue, we investigate the structural properties of Twitter network topologies and propose communitybased features.These community-based features are more robust than profile-based and content-based features due to the fact that community structure is determined by multiple accounts collectively.Compared to a normal user, a spammer is more likely to connect with other spammers.Moreover, we find that spammers often need to fulfill a task in a short period of time to reduce costs.Based on this phenomenon, we design new timebased features to capture spam outbreaks.In addition to effectiveness, we also consider efficiency as another important factor.We select computation-efficient features as potential candidates by comparing the time of feature construction.Our data-driven evaluation demonstrates that our advanced features largely improve the performance of the spam detection system.While achieving an even lower false positive rate, the detection rate increases by 16% and f1 score increases by over 10% when applying our feature set.6 Conclusion and Future Work 6.1 Conclusion . . . . . . . .
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.000 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.005 | 0.000 |
| Scholarly communication | 0.001 | 0.002 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".